supabase / etl / 37741656755

78%

push

github

 clickhouse: retry temporary DDL and lookup errors (#1095)

* clickhouse: retry temporary DDL and lookup errors

The error kind came from which call failed, not from what went wrong.
A dropped connection or a server timeout during DDL or a schema lookup
needed an operator, while the same error during an insert was retried.
A short ClickHouse restart could leave tables and pipelines stuck.

Add is_retryable() on ClickHouse errors. It matches lost connections,
timeouts, a short list of busy or unavailable server codes, and 5xx
statuses. DDL and schema lookups now retry those and still stop on
real rejections. The engine and version lookups go through the same
path and get a timeout.

* clickhouse: don't retry public-address rejections

The previous commit retried every network error. That included the
public-address guard, which rejects a private DNS answer at connect
time with a PermissionDenied I/O error. The rejection repeats on
every attempt, so retrying only delays the stop.

Treat a network error as permanent when its chain holds a
PermissionDenied I/O error. Other connection errors still retry.

122 of 125 new or added lines in 2 files covered. (97.6%)

48593 of 62243 relevant lines covered (78.07%)

13735.03 hits per line